一种基于视觉显著性分析的视频转码算法  

A transcoding algorithm based on visual saliency

作　　者：蒋炜[1] 田翔[1] 陈耀武[1,2] 

机构地区：[1]浙江大学数字技术及仪器研究所,浙江杭州310027 [2]浙江大学浙江省网络多媒体重点实验室,浙江杭州310027

出　　处：《华中科技大学学报（自然科学版）》2014年第4期35-39,共5页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(40927001);浙江科学技术创新团队项目(2011R09021-02)

摘　　要：根据人眼视觉的特点,提出一种基于视觉显著性分析的视频转码算法.首先利用H.264码流中的运动矢量场进行全局运动估计和局部运动分割,然后结合编码比特数的分布特点生成最终的视觉显著性图,最后在转码的重编码过程中,利用视觉显著性图对非显著性区域进行自适应频率系数压制,以在保持视频主观质量的前提下进一步提高压缩效率.实验结果表明:与现有视觉显著性区域检测方法相比,本算法能更准确完整地检测出视频中的显著性区域;而与参考的标准转码算法相比,平均降低约14%的编码比特率,同时保持了几乎相同的主观质量.Transcoding H. 264 video to high efficiency video coding （HEVC） video is important to en- able gradual migration to HEVC. Based on the characteristics of human vision system, a visual sali- ency based video transcoding algorithm from H. 264 to HEVC was proposed. First, the motion vector field included in the H. 264 bitstream was utilized to do the global motion estimation and motion seg- mentation. Second, the distribution of coding bits was combined to produce the visual saliency map which indicated the salient regions in the videos. Finally, during the re-encoding process, a frequency coefficient suppression technique in the transform domain was used in the non-salient region for fur- ther bitrate reduction while keeping the subjective quality of salient region. The experimental results demonstrate that the proposed method outperforms other video saliency detection algorithms and the reference transcoding algorithm, and can reduce the coding bitrate by about 14% under the same sub- jective video quality.

关 键 词：视频转码 视觉显著性 运动矢量场 编码比特数 H 264 高效率视频编码 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]